قیام امن اور مذہبی ہم آہنگی

The Internal dissensions within the ranks of the Muslim Ummah are very harmful and condemnable. Today, the Muslims of the world have fallen into the deep recesses of decline due to their mutual differences. The intrigues and conspiracies of the hostile nations have created schism and dissensions among the Muslims on the grounds of language, land, race and color. In our country (Pakistan), if we ponder on the growing rate of violence, we will find that the main causes of this chaos are our attitude towards our mutual differences. Because of intolerant approach towards our mutual differnces, our difficulties and problems are sizing up, and they have engulfed the whole nation, now. The only point on which our nation can be united is the “Kalimah”. The followers of this “Kalimah” whether they are white or black, rich or poor, or whatever race they belong to, and whatever territory or country they come from, they are all considered as the member of the Muslim Ummah. Keeping the prevailing situation of the Muslim Ummah, the author of this paper feelss that an appropriate answer to the question, ‘are all sorts of differences condemnable?’, is key to end most of our differences. In fact, all sorts of differences are not condemnable or forbidden; if differences of opinions are based on some logical grounds within the jurisdiction of the Qur’ān and Aḥādīth, they are permissible and justified as inevitable and natural. Such kind of approach can promote tolerance and unity among the Muslim Ummah and can put us at peace.

Exploiting Machine Learning Techniques for Cancer Classification in Histopathology

The advancement in microscopic imaging techniques results in the generation of a large amount of high quality data in no time. The accurate, real time and autonomous analysis of this data is crucial for the theoretical biomedical research and clinical diagnosis. A lot of vigorous attempts are dedicated for the evolution of computer aided techniques, which improve human diagnosis by increasing efficiency, decreasing variability in the observations and reducing the human effort on labelling and classifying images. Among such determined attempts, histology image classification is one of the most significant fields due to its extensive application in pathological diagnosis such as tumor/cancer diagnosis. Inherent heterogeneous nature and random spatial intensity differences of histopathology images make the histology tissue classification a complex task. In this thesis four novel, robust and adaptive frameworks are proposed for automated and correct classification of histopathology images. The goal of this research is to achieve the expert pathologists’s diagnosis by catering inherent complexity and prevailing the variations in the opinion of different pathologists. The key contributions of this reseach are: First, a histopathological classification problem is explored from all the perspectives by performing pattern analysis at image level and cell level individually and collectively. The first proposed framework is an abstract feature based framework which performs image-level analysis to capture global texture information. The second framework performs cell-level analysis to get nuclei structure and texture. The third and fourth frameworks perform cell-level and image-level analysis to get nuclei structure and image global texture. Second, the exploration of RGB colour space is preferred to mimic the pathologists’ diagnosis process. The imperative role of RGB color channels is investigated in histopathology image classification by extracting nuclear and global image features across these color channels. Third, instead of analyzing various colour spaces a number of feature measures are explored from spatial and frequency domain to encounter maximum diversity. The individual and combined effect of a large number of statistical, structural and spectral feature measures are analyzed including co-occurrence matrices, run-length matrices, local binary patterns with Fourier transforms, morphology features and intensity features. Fourth, an extensive investigation of rank-based feature selection schemes is performed and proved that elitism is not an optimal strategy for feature selection in histopathology image classification. An abstract feature i.e. an optimal combination of functionally collaborating features having implicit linkages is identified based on classification accuracy through evolutionary search process. Fifth, a number of classifiers are explored and an automatic selection of parameters of classification model (classifier and classifier’s parameters) is performed through Genetic Algorithm based evolutionary technique. The experimentation is performed on images of grade-I benign meningioma four subtypes (meningothelial, fibroblastic, transitional and psammomatous) and pre-invasive breast lesions four classes (usual ductal hyperplasia (UDH) and three nuclear grades of ductal carcinoma in situ (DCIS)). The proposed frameworks achieved the promising classification results for four meningioma subtypes and breast lesions grades. In most of the cases, optimal sets of features obtained from the combination of three color channels and classified through linear support vector machine classifier presented the highest classification accuracy. The extraction of nuclear texture in spatial and frequency domain presented promising classification results.